{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T06:07:37Z","timestamp":1788934057469,"version":"build-2803163510"},"reference-count":47,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T00:00:00Z","timestamp":1788912000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Digit. Health"],"abstract":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>Mental health triage-level intent classification in linguistically diverse and resource-constrained clinical environments requires models that simultaneously meet strict patient data privacy regulations, cross-lingual generalization to low-resource languages, and real-time inference on edge hardware. No existing federated NLP system has jointly demonstrated this combination. This work proposes Fed-XLM-R, a unified federated learning framework designed to close that gap.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>Fed-XLM-R integrates XLM-RoBERTa with lightweight bottleneck adapter layers, adapter-scoped differentially private stochastic gradient descent (DP-SGD), proximal regularization for straggler-robust aggregation (FedProx), and INT8 post-training quantization for edge deployment. Gradient perturbation is confined exclusively to the 0.87% trainable adapter subspace rather than the full 270-million-parameter model. The framework was evaluated in a simulated 10-client federation on a public benchmark, with intent classification trained and evaluated on English data. A cross-lingual diagnostic was additionally run on the XNLI benchmark across five languages (English, Spanish, Hindi, Swahili, Arabic) against an mBERT baseline, and canary phrase memorization analysis was conducted across all 50 federated rounds to assess data extraction risk.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      At the clinically meaningful privacy budget of\n                      <jats:italic>\u03b5<\/jats:italic>\n                      \u2009=\u20091.0 (\n                      <jats:italic>\u03b4<\/jats:italic>\n                      \u2009=\u200910\n                      <jats:sup>\u22125<\/jats:sup>\n                      , Fed-XLM-R achieves 96% accuracy retention, yielding an overall accuracy of 0.923, F1-score of 0.918, and AUC-ROC of 0.956, within 0.5% of a centralized baseline while providing formal differential privacy guarantees. Canary phrase analysis confirms the DP-SGD mechanism prevents training data extraction. The XLM-R backbone retains strong discriminative capacity in low-resource languages relative to mBERT on the XNLI diagnostic. Adapter-only parameter transmission reduces cumulative communication cost by 40.7-fold at the 85% accuracy threshold, at a per-round volume compatible with standard mobile data networks. INT8 quantization reduces inference latency from 1,974 ms to 523 ms on low-end CPU hardware, a 3.8-fold speedup.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>These findings characterize the mechanism's behavior under controlled conditions in a simulated federation rather than demonstrating field deployment. The cross-lingual results support the feasibility of future multilingual extension but do not themselves constitute multilingual intent classification. The three-category label scheme (Anxiety\/Depression\/Normal) supports triage-level intent routing but has not been validated against standardized clinical instruments; the system is not proposed as a clinical diagnostic or screening tool. Overall, Fed-XLM-R demonstrates that privacy-preserving, communication-efficient, and edge-deployable federated NLP is achievable without substantial accuracy loss, offering a practical pathway toward privacy-compliant mental health triage systems pending real-world and multilingual validation.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fdgth.2026.1856286","type":"journal-article","created":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T05:58:39Z","timestamp":1788933519000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Fed-XLM-R: a privacy-preserving federated framework with adapter-scoped differential privacy for mental health triage-level intent classification on resource-constrained edge devices"],"prefix":"10.3389","volume":"8","author":[{"given":"Karthiga","family":"M","sequence":"first","affiliation":[{"name":"Department of Computer Science and Engineering, Bannari Amman Institute of Technology","place":["Sathyamangalam, Erode, Tamilnadu, India"]},{"name":"Postdoctoral Researcher, Multimedia University","place":["Melaka, Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Emerson Raja","family":"Joseph","sequence":"additional","affiliation":[{"name":"Faculty of Engineering and Technology, Centre for Advanced Analytics COE of Artificial Intelligence, Multimedia University","place":["Melaka, Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Subhash","family":"Patil","sequence":"additional","affiliation":[{"name":"Professor and Head of Surgery Department, International Medical University, Management and Science University","place":["Federal Territory of Kuala Lumpur, Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"K.","family":"Saranya","sequence":"additional","affiliation":[{"name":"Department of Computer Science and Engineering, Bannari Amman Institute of Technology","place":["Sathyamangalam, Erode, Tamilnadu, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,9]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"28","DOI":"10.1007\/s42044-025-00375-0","article-title":"NLP-driven federated learning framework for multi-organ rare disease prediction","volume":"9","author":"Kunjumohamad","year":"2026","journal-title":"Iran J Comput Sci"},{"key":"B2","doi-asserted-by":"publisher","first-page":"111514","DOI":"10.1016\/j.compbiomed.2026.111514","article-title":"Ollama-driven medical insights using LLMs with a federated learning approach","volume":"204","author":"Lal","year":"2026","journal-title":"Comput Biol Med"},{"key":"B3","doi-asserted-by":"publisher","first-page":"921","DOI":"10.1109\/TASLPRO.2026.3657644","article-title":"LLM-driven synthetic text generation for privacy-preserving federated learning","volume":"34","author":"Ma","year":"2026","journal-title":"IEEE Trans Audio Speech Lang Process"},{"key":"B4","doi-asserted-by":"publisher","first-page":"113527","DOI":"10.1016\/j.engappai.2025.113527","article-title":"Federated learning with transient central initialization and decentralized edge aggregation","volume":"165","author":"Nanor","year":"2026","journal-title":"Eng Appl Artif Intell"},{"key":"B5","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1109\/TCSS.2025.3638431","article-title":"MPCPFL: prototype-based multimodal personalized federated learning with missing modalities in medical domain","volume":"13","author":"Yu","year":"2026","journal-title":"IEEE Trans Comput Soc Syst"},{"key":"B6","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1080\/15265161.2026.2637093","article-title":"Federation opacity and the promise of federated learning in healthcare","volume":"26","author":"Hatherley","year":"2026","journal-title":"Am J Bioeth"},{"key":"B7","doi-asserted-by":"publisher","first-page":"78","DOI":"10.64751\/ijdim.2026.v5.n1.pp78-88","article-title":"Privacy-preserving generative AI in healthcare systems using federated learning approaches","volume":"5","author":"Poojari","year":"2026","journal-title":"Int J Data Sci IoT Manag Syst"},{"key":"B8","doi-asserted-by":"publisher","first-page":"108576","DOI":"10.1016\/j.bspc.2025.108576","article-title":"Federated learning for melanoma detection: a decentralized and privacy-preserving approach using deep learning","volume":"112","author":"Din","year":"2026","journal-title":"Biomed Signal Process Control"},{"key":"B9","doi-asserted-by":"publisher","first-page":"9768","DOI":"10.1038\/s41598-026-40581-9","article-title":"Privacy-preserving federated learning with light-weight attention improved CNNs for automated leukemia detection across distributed medical imaging","volume":"16","author":"Awan","year":"2026","journal-title":"Sci Rep"},{"key":"B10","doi-asserted-by":"publisher","first-page":"55","DOI":"10.1007\/s10115-025-02673-2","article-title":"A comprehensive survey on federated learning for privacy preservation in digital healthcare applications","volume":"68","author":"Kanauzia","year":"2026","journal-title":"Knowl Inf Syst"},{"key":"B11","doi-asserted-by":"publisher","first-page":"8","DOI":"10.54097\/d52m6j10","article-title":"Federated learning approaches for privacy-preserving big data analytics","volume":"20","author":"Kou","year":"2026","journal-title":"J Comput Electron Inf Manag"},{"key":"B12","doi-asserted-by":"publisher","first-page":"109557","DOI":"10.1016\/j.bspc.2026.109557","article-title":"Federated learning with homomorphic encryption for secure real time ECG anomaly detection: a multi institutional privacy preserving framework","volume":"116","author":"Yuksel","year":"2026","journal-title":"Biomed Signal Process Control"},{"key":"B13","doi-asserted-by":"publisher","first-page":"108426","DOI":"10.1016\/j.bspc.2025.108426","article-title":"PriFL-XAI: hybrid privacy-preserving federated learning models for monkeypox detection through GAN augmentation and explainable AI","volume":"112","author":"Mahir","year":"2026","journal-title":"Biomed Signal Process Control"},{"key":"B14","first-page":"1","article-title":"Model stability under client diversity: a comparative study of federated aggregation algorithms in bangla NLP","author":"Julekha","year":"2026"},{"key":"B15","doi-asserted-by":"publisher","first-page":"108474","DOI":"10.1016\/j.comcom.2026.108474","article-title":"Per-FedDMA: a personalized federated learning method based on deep multisampling and hypernetwork dynamic adaptation","volume":"251","author":"Liu","year":"2026","journal-title":"Comput Commun"},{"key":"B16","doi-asserted-by":"publisher","first-page":"132972","DOI":"10.1016\/j.neucom.2026.132972","article-title":"FedACA: adaptive classifier aggregation and clustering for personalized heterogeneous federated learning","volume":"674","author":"Dong","year":"2026","journal-title":"Neurocomputing"},{"key":"B17","doi-asserted-by":"publisher","first-page":"11940","DOI":"10.1109\/JSTARS.2026.3670968","article-title":"Communication-efficient federated swarm learning for SAG-CM on heterogeneous data","volume":"19","author":"Xu","year":"2026","journal-title":"IEEE J Sel Top Appl Earth Obs Remote Sens"},{"key":"B18","first-page":"1","article-title":"Optimizing federated learning with particle swarm intelligence for next-generation video surveillance","author":"Newegy","year":"2026"},{"key":"B19","doi-asserted-by":"publisher","first-page":"1433","DOI":"10.1109\/TC.2026.3654074","article-title":"Communication-efficient federated learning by exploiting spatio-temporal correlations of gradients","volume":"75","author":"Zheng","year":"2026","journal-title":"IEEE Trans Comput"},{"key":"B20","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1109\/JBHI.2026.3676321","article-title":"FedGA: genetic algorithm-guided federated learning for medical image segmentation with non-IID features","volume":"20","author":"Ahmed","year":"2026","journal-title":"IEEE J Biomed Health Inform"},{"key":"B21","first-page":"0620","article-title":"DP-Prox: a robust and differentially private framework for federated instruction tuning of small LLMs on 8 GB edge devices","author":"Venkatesh","year":"2026"},{"key":"B22","doi-asserted-by":"publisher","first-page":"1406","DOI":"10.3390\/en19061406","article-title":"Federated distributed scheduling for hydrogen production under renewable variability: a safety-constrained evaluation of FedAvg, FedProx, gossip, and local control","volume":"19","author":"Al-Shammari","year":"2026","journal-title":"Energies"},{"key":"B23","doi-asserted-by":"publisher","first-page":"12953","DOI":"10.1038\/s41598-026-42786-4","article-title":"TigCLaF: a cross-lingual large language model framework for sentiment-aware text classification in low-resource tigrigna","volume":"16","author":"Gebremeskel","year":"2026","journal-title":"Sci Rep"},{"key":"B24","doi-asserted-by":"publisher","first-page":"e70648","DOI":"10.1002\/cpe.70648","article-title":"News classification in low-resource languages: insights from transformer and baseline models","volume":"38","author":"Demilie","year":"2026","journal-title":"Concurrency Comput Pract Exper"},{"key":"B25","doi-asserted-by":"publisher","first-page":"47","DOI":"10.1007\/s13278-026-01588-2","article-title":"Sentiment analysis for code-mixed low-resource languages: a systematic review of approaches, techniques, applications, challenges, and future directions","volume":"16","author":"Nazir","year":"2026","journal-title":"Soc Netw Anal Min"},{"key":"B26","doi-asserted-by":"crossref","first-page":"173","DOI":"10.1007\/978-3-031-96265-3_12","article-title":"Federated learning: a paradigm shift in healthcare data privacy","volume-title":"Integrating Cloud, Fog, and Edge Computing in Healthcare: Federated Learning and Blockchain Approaches: Harnessing Distributed Technologies for Enhanced Healthcare Delivery","author":"Goel","year":"2026"},{"key":"B27","first-page":"1","article-title":"Transfer learning-driven adaptive BERT model for masked language modeling in Indigenous languages","author":"Biswas","year":"2026"},{"key":"B28","first-page":"1","article-title":"Multilingual NLP models: a review of techniques, tools, and real-world use cases","author":"Kasilingam","year":"2026"},{"key":"B29","first-page":"453","article-title":"Kyrtext: a multi-domain large-scale corpus for Kyrgyz language","author":"Chubakov","year":"2026"},{"key":"B30","doi-asserted-by":"publisher","first-page":"2511","DOI":"10.1007\/s41870-025-03055-1","article-title":"A hybrid deep learning framework for multilingual depression detection and symptom classification from social media text","author":"Shwetha","year":"2026","journal-title":"Int J Inf Technol"},{"key":"B31","first-page":"1496","article-title":"Contextual sentiment analysis using transformer-based models","author":"Pyingkodi","year":"2026"},{"key":"B32","first-page":"1","article-title":"Comparative analysis of depression comment classification using traditional machine learning and transformer-based models with easy data augmentation","author":"Lahuddin","year":"2026"},{"key":"B33","doi-asserted-by":"publisher","first-page":"2640018","DOI":"10.1142\/S1756973726400184","article-title":"Transformer-based multilingual NLP model for low-resource language translation","volume":"17","author":"Alghamdi","year":"2026","journal-title":"J Multiscale Modell"},{"key":"B34","doi-asserted-by":"publisher","first-page":"191","DOI":"10.1038\/s44360-025-00038-1","article-title":"Large language models for frontline healthcare support in low-resource settings","volume":"1","author":"Rutunda","year":"2026","journal-title":"Nat Health"},{"key":"B35","doi-asserted-by":"publisher","first-page":"306","DOI":"10.3390\/healthcare14030306","article-title":"Federated learning in healthcare ethics: a systematic review of privacy-preserving and equitable medical AI","volume":"14","author":"Mir","year":"2026","journal-title":"Healthcare"},{"key":"B36","doi-asserted-by":"publisher","first-page":"131501","DOI":"10.1016\/j.eswa.2026.131501","article-title":"Roughness-informed differential privacy","volume":"313","author":"Partohaghighi","year":"2026","journal-title":"Expert Syst Appl"},{"key":"B37","doi-asserted-by":"publisher","first-page":"108502","DOI":"10.1016\/j.bspc.2025.108502","article-title":"Securing deep learning models with differential privacy for cardiovascular disease prediction","volume":"112","author":"Orabe","year":"2026","journal-title":"Biomed Signal Process Control"},{"key":"B38","first-page":"179","article-title":"Differential privacy and homomorphic encryption in healthcare artificial intelligence","author":"Shafik","year":"2026"},{"key":"B39","doi-asserted-by":"publisher","first-page":"114586","DOI":"10.1016\/j.asoc.2026.114586","article-title":"An enhanced privacy-preserving federated few-shot learning framework for respiratory disease diagnosis","volume":"190","author":"Wang","year":"2026","journal-title":"Appl Soft Comput"},{"key":"B40","doi-asserted-by":"publisher","first-page":"3242","DOI":"10.1109\/TON.2026.3658317","article-title":"Privacy-utility trade-off in federated LLM fine-tuning: a dynamic game approach","volume":"34","author":"Wang","year":"2026","journal-title":"IEEE Trans Netw"},{"key":"B41","doi-asserted-by":"publisher","first-page":"5462","DOI":"10.1109\/TDSC.2026.3655785","article-title":"Sleight: hidden data privacy breaches in federated learning","volume":"23","author":"Gong","year":"2026","journal-title":"IEEE Trans Dependable Secure Comput"},{"key":"B42","doi-asserted-by":"publisher","first-page":"26","DOI":"10.69987\/AIMLR.2026.70102","article-title":"Adaptive privacy budget allocation for differential privacy optimization in fleet federated learning: algorithm enhancement and performance evaluation","volume":"7","author":"Guo","year":"2026","journal-title":"Artif Intell Mach Learn Rev"},{"key":"B43","doi-asserted-by":"publisher","first-page":"109312","DOI":"10.1016\/j.bspc.2025.109312","article-title":"Demography-aware personalized federated learning for fair, private, and efficient clinical risk prediction","volume":"114","author":"Zhao","year":"2026","journal-title":"Biomed Signal Process Control"},{"key":"B44","first-page":"143","article-title":"Weighted federated learning with encryption for diabetes classification","author":"Zhao","year":"2025"},{"key":"B45","doi-asserted-by":"publisher","first-page":"181","DOI":"10.1007\/s41666-024-00164-7","article-title":"Analyzing the impact of personalization on fairness in federated learning for healthcare","volume":"8","author":"Wang","year":"2024","journal-title":"J Healthcare Inf Res"},{"key":"B46","first-page":"2790","article-title":"Parameter-efficient transfer learning for NLP","author":"Houlsby","year":"2019"},{"key":"B47","first-page":"2704","article-title":"Quantization and training of neural networks for efficient integer-arithmetic-only inference","author":"Jacob","year":"2018"}],"container-title":["Frontiers in Digital Health"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fdgth.2026.1856286\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,9]],"date-time":"2026-09-09T05:58:40Z","timestamp":1788933520000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fdgth.2026.1856286\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,9]]},"references-count":47,"alternative-id":["10.3389\/fdgth.2026.1856286"],"URL":"https:\/\/doi.org\/10.3389\/fdgth.2026.1856286","relation":{},"ISSN":["2673-253X"],"issn-type":[{"value":"2673-253X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,9]]},"article-number":"1856286"}}